Applied Thermal Engineering 125 (2017) 377–385

Contents lists available at ScienceDirect

Applied Thermal Engineering

journal homepage: www.elsevier.com/locate/apthermeng

Research Paper

Simultaneous optimization of pump and cooler networks in a cooling

water system
Jiaze Ma a, Yufei Wang a,⇑, Xiao Feng b
a
State Key Laboratory of Heavy Oil Processing, China University of Petroleum, Beijing 102249, China
b
School of Chemical Engineering & Technology, Xi’an Jiaotong University, Xi’an 710049, China

h i g h l i g h t s

Series-parallel cooler network is proposed to reduce total flowrate of cooling water in cooling water system.

Considering installation heights of coolers and employing main-auxiliary pump network to reduce pumping cost.

Series-parallel cooler network and main-auxiliary pump network are optimized simultaneously.

Interconnections between pump network and cooler network have been analyzed.

The effect of cooler installation heights on system is investigated.

a r t i c l e i n f o a b s t r a c t

Article history: A cooling water system requires a large amount of energy to transport the cooling water. It contains two
Received 24 March 2017 subsystems, namely, a cooler network and a pump network. Conventionally, these two subsystems are
Revised 12 June 2017 optimized separately. The intrinsic connections between the pump cost, cooler cost, and cooling water
Accepted 3 July 2017
flow rate, require the cooler and pump networks to be optimized simultaneously. This paper presents
Available online 5 July 2017
an optimization model for a cooling water system in which both the subsystems are simultaneously opti-
mized. For the cooler network, a series–parallel superstructure is employed to reduce the cooling water
Keywords:
flow rate for decreasing the pumping cost. A main–auxiliary pumps structure is applied in the pumping
Cooling water system
Pump network
system for reducing the energy consumption. The intrinsic relationship between the cooler network and
Cooler network pumping system is investigated. The model is formulated as a mixed-integer nonlinear programming
MINLP (MINLP) problem. The objective is to determine the cooling water system by minimizing the total annual
cost. A case study is employed to demonstrate the effectiveness of the proposed model. The results show
that the simultaneous optimization method can lead to a 6.3% TAC reduction in comparison with the two-
step sequential optimization method.
Ó 2017 Elsevier Ltd. All rights reserved.

1. Introduction
Cooling water system is widely used to remove waste heat in
large industrial facilities, such as petroleum refineries, chemical
plants, and power stations. Because water has suitable thermal
properties and non-harmful chemical composition, cooling water
system, by far, is used as the most common cold utility in industry.
The two most important components in a cooling water system are
cooler network and pump network. According to previous
research, pumping systems, particularly centrifugal pumps, con-
sume approximately 20% of the total energy of the world [1].
⇑ Corresponding author.
E-mail address: wangyufei@cup.edu.cn (Y. Wang).
Designing optimal cooler and pump networks can help reduce
the pump energy consumption.
In the last few years, several optimization models have been
proposed for designing suitable cooler networks. Kim and Smith
[2] introduced a mathematical model that focuses on minimizing
the total water flow rate. Cooling water can be reused in a mixed
parallel/series cooler arrangement, and the cooling tower exhibits
a better performance owing to the reduction in the cooling water
flow rate and increase in the cooling water return temperature.
Feng et al. [3] proposed a cooling water network with an interme-
diate main, which was easy to control and operate. By retrofitting
the cooling water system with an intermediate main, the cooling
water flow rate decreased, and the cooling tower effectiveness
increased. Panjeshahia et al. [4] combined pinch technology and
mathematical programming for minimizing the cooling water

http://dx.doi.org/10.1016/j.applthermaleng.2017.07.026
1359-4311/Ó 2017 Elsevier Ltd. All rights reserved.
378 J. Ma et al. / Applied Thermal Engineering 125 (2017) 377–385
Nomenclature
a, b, c constants of exchanger capital cost
Af annualized factor
Ar(i) exchanger area of cooler E-i
cp specific heat capacity of cooling water
Cfpump fixed charge of pumps
Cpump pressure cost coefficient for pumps
Ccu unit cost of cooling water
di internal tube diameter
dt external tube diameter
dtin(i) temperature approach between hot stream outlet
temperature and cold stream inlet temperature
dtout(i) temperature approach between hot stream inlet
temperature and cold stream outlet temperature
e unit cost of electricity
fw(i) mass flow rate of fresh cooling water in cooler E-i
fin(i) mass flow rate of cooling water in cooler E-i
gc gravitational constant
h annual operational time
h(i) film transfer coefficient of hot stream i
hw film transfer coefficient of cooling water
H(i) minimum pressure head of cooler E-i
Ha(i) pressure head of auxiliary pump i
Hdi,j installation height of E-i minus installation height of E-j
positiv e
Hdi;j positive value of Hdi,j
installh(i) installation height of cooler E-i
system cost. The authors even considered the environmental
impact, except for water and energy conservation. It is to be noted
that all the models previously mentioned mainly focus on mini-
mizing the water flow rate. However, to achieve heat exchanging
service, more contact areas are required. Subsequently, Ponce-
Ortega et al. [5] proposed a mixed-integer nonlinear programming
(MINLP) model that considered the capital cost of the coolers and
cooling water cost simultaneously. In their work, a stage-wise
cooler network superstructure was employed. The objective was
to minimize the total cost, and this model was more economical
compared with previously mentioned models. However, in a cool-
ing water system, a stage-wise superstructure is less practical
compared with a series–parallel structure. The pressure drop of
the system is a sum of the pressure drops of all the stages; There-
fore, when coolers are arranged in three or more stages, owing to
the large pressure drop, the pumping cost can be extremely high.
Furthermore, more splitters and mixers are required in a stage-
wise superstructure. Wang et al. [6] formulated a two-step method
to convert a parallel configuration into a series–parallel structure
without requiring additional contact area. Modifying the configu-
ration into a series–parallel structure, significantly reduced the
water consumption, and thereby, decreased the pumping cost.
Converting a parallel cooler network into a series–parallel arrange-
ment increases the possibility of water reusability. However, cool-
ers in series have a higher pressure drop. Picón-Núñez et al. [7]
studied the effect of the cooler network arrangement on the cool-
ing water system. They considered the pressure drop effect and
pumping cost in the optimization of the cooler network under
the constraints of permissible water return temperature and pres-
sure drop. Their model can significantly reduce the operational
cost of existing systems. Besides converting a parallel network into
series–parallel structure, adding a new cooler to an existing system
is an alternative approach to decrease the cost. Picón-Núñez et al.
[8] have proposed reasonable methods to identify the location for a
new cooler by considering the thermal–hydraulic performance of
the system. The proposed method guides engineers in adding a
kt conductivity of cooling water
Kt(i) parameter used in relating physical properties to pres-
sure drop
OCpump total pump operational cost
OCmainpump main pump operational cost
OCauxipump
auxiliary pump operational cost
CCpump pump capital cost
pt(i) cooler E-i tube pressure drop
4ptotal total pressure drop of cooler network
q(i) heat load of cooler E-i
Thin(i) hot stream inlet temperature of cooler E-i
Thout(i) hot stream outlet temperature of cooler E-i
4Tmin minimum approach temperature difference
Tfw temperature of fresh cooling water
yi binary variable deciding whether an auxiliary pump ex-
ists
Greek Letters
Ut viscosity correction factor
mt cooling water viscosity
g pump efficiency
q density of cooling water
c exponent for the pump cost function
new cooler to the cooling water system properly. Li et al. [9]
employed three water mains in their cooler network. The proposed
superstructure covered majority of the possible network connec-
tions. This model increased the opportunities for water reuse and
improved the flexibility of the cooling water system.
Besides the flow rate of cooling water, pump energy consump-
tion also depends on the pump pressure head. The pump pressure
head is related to installation height of the cooler and system pres-
sure drop. The pumping cost constitutes a major part of the total
cost of a cooling water system, optimizing a cooling water system
by minimizing the pump pressure head is an effective method to
save cost. Shankar et al. [10] used the variable frequency drives
control technique to improve the efficiency of pumping. Tirmizi
et al. [11] substituted the existing single-loop constant pumping
scheme by a constant primary and variable secondary scheme. This
configuration saved up to 8% of the energy consumption.
Westerlund et al. [12] introduced a superstructure-based model
to optimize the pump configurations. Zhang et al. [13] studied
the pump configuration under different working conditions in a
wastewater plant. However, these works only focused on a detailed
design of the pump; Optimization for reducing the pressure head
requirement was not considered. The pressure drop in a system
and cooler installation height are two decisive factors for deter-
mining the pump pressure head.
For considering the pressure drop in a system, Frausto-
Hernández et al. [14] optimized the heat exchanger network using
the MINLP model. Another model was proposed by Souza et al. [15]
that simultaneously took into account the layout of the equipment
and pressure drop of the exchangers. Considering the equipment
layout and piping cost in their heat exchanger network yielded a
significant reduction in the total cost. Torkfar and Avami [16]
optimized the water and energy network by accounting for the
pressure drop. The model reflected the interactions between the
thermal and electrical energy and water requirements. In addition,
Hung and Kim [17] considered the relationship between the pres-
sure drop and pumping cost, and explored the trade-offs between
 J. Ma et al. / Applied Thermal Engineering 125 (2017) 377–385
the piping, pumping, and water costs. Gololo and Majozi [18] sug-
gested a complex cooling water system by including the pressure
drop effect. The proposed technique offered an opportunity to
debottleneck cooling water systems with multiple cooling towers
while maintaining a minimum pressure drop. Soltani and Shafiei
[19] constructed a model that optimized the pressure drop using
coupling algorithms. In their work, the pressure drops of the
streams were calculated from the results of an LP, and subse-
quently, a modified ILP problem was solved to determine the max-
imum profit of retrofitting the HENs.
Installation heights of the coolers also affect the pump pressure
head; however, because the installation height is a fixed parame-
ter, this parameter does not receive significant attention. Recently,
Sun et al. [20] have proposed a novel pump superstructure that can
be implemented to minimize the pumping cost. This pump scheme
involves a main pump and several auxiliary pumps that are
installed on the branch pipelines. The heights of the coolers in
the cooling water network are varied and the main pump pressure
head is required to be higher than minimum pressure heads
requirement of all the coolers. By installing auxiliary pumps on
the branch pipelines, the main pump pressure head reduces and
consequently, the cost of the main pump decreases. A pumping
system with auxiliary pumps was adopted in this study.
In a cooling water system, the cooler and pump networks inter-
act with each other. Increasing the flow rate of the cooling water
reduces the exchanger area and its capital cost. However, trans-
porting a large amount of water requires more pumping cost, par-
ticularly when the cooler is located at an extremely high position.
Therefore, the cooling water flow rate, exchanger capital cost, and
pumping cost should be considered simultaneously. Sun et al. [21]
proposed a two-step sequential method to optimize the cooler and
pump networks sequentially. The cooler network was optimized
first. With a specified cooler network and water flow rate, the
pump network was optimized sequentially. However, the final
configuration was not optimal because the interconnections
between the pump and cooler and between the pump and water
flow rate were not fully addressed. Ponce-Ortega et al. [22] opti-
mized the cooler network, pumping system, and cooling tower
simultaneously. However, in their work, the pumping cost was
exclusively related to pressure drops of the coolers. The impact
of the installation heights of the coolers on the pumping system
was not considered.
In view of the above background, this study proposes a method
for constructing an optimal cooling water configuration. The objec-
tive of this study is to determine the optimal cooling water config-
uration by optimizing the pump and cooler networks
simultaneously. A series–parallel cooler network has a low pres-
sure drop and high flexibility, and it is more practical in compar-
ison with a stage-wise superstructure; Therefore, it is employed
in this study. A pump network with main and auxiliary pumps is
implemented. By installing the auxiliary pumps on the branch
pipelines, the total operational cost of the pumps is reduced. The
pressure drop has a significant impact on the total cost; therefore,
the pressure drop is treated as an optimized variable. Furthermore,
the installation height of the cooler is also considered. The problem
is formulated as an MINLP model. The intrinsic interactions
between the cooler and pump networks are explored. An example
from Sun et al. [21] is employed to demonstrate the effectiveness
of this model.
2. Problem statement
The intrinsic interactions between the water flow rate, heat
exchanger capital cost, and pumping cost requires the cooler and
pump networks to be optimized simultaneously. Given a set of
379
hot streams with known flowrates, inlet and outlet temperatures,
and film transfer coefficients, the series–parallel cooler network
was employed to reduce the cooling water consumption and
pumping cost. A pump network with auxiliary pumps was pro-
posed to minimize the main pump pressure head and reduce the
pumping cost. The objective was to optimize the cooler and pump
networks simultaneously to decrease the power consumption. The
variables included water flow rates of different coolers, exchanger
areas, pump pressure heads, and pressure drops between the cool-
ers. The aim of this study was to determine the optimal structure
that satisfied the required heat exchanging service with a mini-
mum total annual cost. The MINLP model was employed to solve
the problem, and the optimal design was based on the following
several assumptions:
(1). The specific heat capacities and film transfer coefficients of
the hot and cold streams were constants.
(2). Each hot stream corresponded to only one cooler.
(3). The coolers were 1-1 counter current shell and tube
exchangers.
3. Model formulation
3.1. Series–parallel superstructure model
Fig. 1 shows the inlet and outlet streams of each cooler unit. In
the series–parallel structure, each cooler inlet cooling water
stream is either the fresh cooling water from the cooling tower
or the used cooling water from another cooler. The outlet cooling
water is sent to either the cooling tower or another cooler for
reuse. There are several constraints that illustrate the relationships
between the inlet and outlet streams. In Eq. (1), binary variable Z(i,
j) represents whether cooler E-i sends the outlet cooling water to
cooler E-j. When Z(i,j) is 1, there is a connection between coolers
E-i and E-j. Each cooler, e.g., cooler E-i, can only send the outlet
cooling water to only a single cooler or return the cooling water
directly to the cooling tower.
X
n
Zi; j  1 ð1Þ
j¼1
Each cooler, e.g., cooler E-j, can only accept the used cooling
water from only a single cooler or receive fresh cooling water from
the cooling tower. This relationship is represented in Eq. (2).
X
n
Zi; j  1 ð2Þ
i¼1
The intention of introducing the above constrains is to simplify
the cooler network. If one cooler connects with multiple coolers,
then more splitters and mixers are required, and a very complex
structure is undesirable in practical applications in the industry.
In addition, the coolers cannot reuse their own outlet cooling
water. Eq. (3) expresses this relationship, i.e., when i = j, Z(i,j) is 0.
Zi; j ¼ 0; i ¼ j: ð3Þ
To define the inlet temperature and mass flow rate of each
cooler, Eq. (4) and Eq. (5) are employed.
" #
X
n X
n
TinðiÞ ¼ T fw  1  Z j;i þ ðZ j;i  ToutðjÞÞ ð4Þ
j¼1 j¼1
" #
X
n X
n
finðiÞ ¼ fwðiÞ  1  Z j;i þ ðZ j;i  fwðjÞÞ ð5Þ
j¼1 j¼1
In Eq. (4) and Eq. (5), fw(i) represents the mass flow rate of the
fresh cooling water in cooler E-i, fin(i) is the mass flow rate of the
380 J. Ma et al. / Applied Thermal Engineering 125 (2017) 377–385

Inlet hot stream

Fresh cooling Send back to
water cooling tower

OR OR

Outlet cooling
water from Outlet hot stream Send to cooler k
cooler i

Fig. 1. Inlet and outlet stream details of each cooler unit.

cooling water in cooler E-i, Tin(i) and Tout(i) are the inlet and outlet side. In this work, only the tube side pressure drop is considered,
temperatures of cooler E-i, respectively, and Tfw is the temperature and Eqs. (12–14) are employed for the tube side pressure drop
of the fresh cooling water. When no used water is sent to cooler E-i, formulation.
the inlet cooling water stream is from the cooling tower and the 3:5
inlet cooling water temperature is equal to the fresh water temper- pt ðiÞ ¼ KtðiÞ  ArðiÞ  ht ð12Þ
ature. When cooler E-i accepts the used water from another cooler Constant Kt is related to cooling water viscosity mt, viscosity cor-
E-j, E-i cannot receive fresh water and the inlet water temperature rection factor Ut, tube internal diameter di, tube external diameter
is the E-j outlet water temperature. Furthermore, when cooler E-i dt, density of water q, conductivity kt, and specific heat capacity cpt.
does not receive water from another cooler, it consumes fresh
 l11=6
0:5
water. However, when cooler E-i receives the used cooling water /4:5  di di
KtðiÞ ¼ t
 ð13Þ
 finðiÞ  qt  kt
2:5 7=3 dt
from another cooler E-j, it will not consume fresh water and the 0:023  C 7=6
pt
flow rate of cooler E-i is equal to flow rate that of E-j.
The total mass flow rate of the cooling water is the sum of all When two coolers are in series, the pressure drop of this branch is
the fresh water mass flow rates sent from the cooling tower to the sum of the pressure drops of the two coolers. For example, when
the coolers. cooler E-i connects with cooler E-j, the pressure drop of this branch
is the sum of the pressure drops of cooler E-i and cooler E-j. The ser-
X
n
ies–parallel structure causes the total pressure drop of the cooler
ft ¼ fwðiÞ ð6Þ
i¼1
network to be the largest pressure drop among the different
branches. When the cooling system is in a stable working condition,
Eq. (7) defines the relationship between the heat load, inlet and out- all the pressure drops of the branch pipelines are the same, and they
let temperatures, and cooling water flow rate in cooler E-i. are all equal to the total pressure drop. Eq. (14) is employed to cal-
qðiÞ ¼ ðToutðiÞ  TinðiÞÞ  cp  finðiÞ ð7Þ culate the total pressure drop of the system.

The calculation of the exchanger area expressed in Eq. (8) is X

based on the Chen approximation for the logarithmic mean of tem-
perature [23]. In Eq. (8), dtin(i) is the temperature approach
between the hot stream outlet temperature and cold stream inlet
temperature, dtout(i) is the temperature approach between the
hot stream inlet temperature and cold stream outlet temperature,
h(i) denotes the film transfer coefficient of hot stream i, and hw is
the film transfer coefficient of the cooling water.
 
qðiÞ 1 1
ArðiÞ ¼ h i1=3  þ ð8Þ
hðiÞ hw
dtinðiÞ  dtoutðiÞ  dtinðiÞþdtoutðiÞ
2
The temperature approaches between cold and hot streams are
required to be larger than the minimum temperature approach dif-
ference. The cooling water outlet temperature should be higher
than the inlet temperature. These constraints are represented in
Eq. (9–11), where Thout(i) and Thin(i) are the hot stream inlet
and outlet temperature, respectively.
dtinðiÞ ¼ ThoutðiÞ  TinðiÞ P DT min
dtoutðiÞ ¼ ThinðiÞ  ToutðiÞ P DT min
ToutðiÞ P TinðiÞ
The tube side pressure drops between the coolers are treated as
optimized variables. The pressure drop is related to constant Kt,
contact area, and film transfer coefficient [19]. Generally, the cool-
ing water flows in the tube side and hot streams flow in the shell
n X
n
Dptotal P ptðiÞ þ ðZ j;i  ptðjÞÞ þ ðZ i;j  ptðjÞÞ ð14Þ
j¼1 j¼1
3.2. Pump network formulation
In this study, auxiliary pumps are introduced in the system. To
transport cooling water to all the coolers, the pressure head of the
main pump is required to be larger than minimum pressure head
requirements of all the coolers. When coolers are installed on an
extremely high platform, the pumping cost is extremely high and
a novel pump structure that was proposed by Sun et al. [20] can
reduce the pump operational cost. By installing the auxiliary
pumps on the specified branches, the pressure head of the main
pump is not required to be larger than the minimum pressure head
requirements of all the coolers; This in turn leads to a decrease in
the main pump pressure head and power cost.
3.2.1. Minimum pressure head requirements of coolers
ð9Þ
The minimum pressure head requirements of the coolers
depend on the installation heights of the coolers and pressure drop
ð10Þ
of the system. When coolers E-i and E-j are arranged in series, the
minimum pressure head requirements of both the coolers are
ð11Þ
based on the highest cooler and pressure drop. However, the cooler
series arrangements are unknown before the optimal configuration
is obtained. In this case, it is necessary to determine the height dif-
ferences between the different coolers. When two coolers are con-
nected to each other, using the known installation height of one
 J. Ma et al. / Applied Thermal Engineering 125 (2017) 377–385 381

!
cooler, height difference between the two connected coolers, and X
n
Ha ðiÞ  fwðiÞ  g
OC auxipump ¼ yi eh ð22Þ
pressure drop of the system, the minimum pressure head require-
i¼1
gpump
ments can be evaluated.Eq. (15) and Eq. (16) are employed to
obtain the height differences between the coolers. In these equa- The objective of the problem is to minimize the total annual
tions, Hdi,j represents the height of cooler E-i minus that of cooler cost that includes the capital and operational costs of the pumps,
E-j. When cooler E-i is lower than cooler E-j, the value of Hdi,j is exchanger capital cost, and cooling water cost.
positiv e 
negative. When Hdi,j is negative, Hdi;j is set to zero. When cooler c
TAC ¼ Af  Cfpump þ C pump ðft  Hm  gÞ
E-i is higher than cooler E-j, the value of Hdi,j is positive. When Hdi,j
positiv e
X
n
 c 
is positive, Hdi;j equals Hdi,j. þ Af  Cfpump yi þ C pump ðfwðiÞ  Ha ðiÞ  gÞ Þ
i¼1
Hdi;j ¼ installhðiÞ  installhðjÞ ð15Þ X
n
fwðiÞ  Ha ðiÞ  g
positiv e ft  Hm  g
Hdi;j ¼ Maxð0; Hdi;j Þ ð16Þ þ ehþ i¼1
eh
gpump gpump
The intention of introducing the height difference is to deter- X
n
 
mine the minimum pressure head requirements of the different þ Af a þ b  Arc ðiÞ þ ft  h  C cu ð23Þ
coolers. H(i) is the minimum pressure head requirement of cooler i¼1

E-i. In Eq. (17), the minimum pressure head requirement depends

on the sum of the cooler installation height and total pressure drop
4. Solution technique employed
of the system. When cooler E-i sends the used cooling water to
cooler E-j, their minimum pressure head requirements are the
In this study, the MINLP model was employed. The model was
same and based on the highest cooler. For example, when cooler
positiv e implemented in software GAMS, and solver DICOPT was used to
E-i is higher than cooler E-j, the value of Hdj;i is zero. When
solve the MINLP problem. No global optimal solution was ensured
positiv e
cooler E-j is higher than the cooler E-i, the value of Hdj;i is their by using DICOPT in this work. Although the obtained result was a
positive height difference. Then, the minimum pressure head local optimum, it was in the vicinity of the global optimum. There-
requirement of cooler E-i is the combination of cooler E-i installa- fore, the obtained result was used to obtain physical insights of this
tion height, positive height difference between the two coolers, work.
and system pressure drop. In addition, when cooler E-i receives
water from cooler E-j and cooler E-j is higher than cooler E-i, the 5. Case study
positiv e
value of Hdj;i is their positive height difference. When cooler
positiv e To verify the effectiveness of the simultaneous optimization, a
E-i is higher than cooler j, the value of Hdj;i is set to zero. Eq.
case study performed by Sun et al. [21] was employed. The hot
(17) is employed to express this relationship.
n 
stream data is shown in Table 1.
X  n 
X 
positiv e positiv e Each hot stream corresponds to only a single cooler, and the
HðiÞ ¼ installhðiÞ þ Zi; j  Hdj;i þ Zj; i  Hdj;i
j¼1 j¼1 installation heights of all the coolers are known. The physical prop-
erties and constraints of the case study are listed in Table 2.
þ Dptotal =q=g ð17Þ
All the coolers are assumed to be 1-1 counter current shell and
tube exchangers, and the cooling water flows along the tube side.
3.2.2. Pressure heads of pumps With the physical properties, the pressure drop constant Kt can
The combination of the main pump pressure head Hm and aux- be determined, whereas the pressure drops of the coolers are
iliary pump pressure head Ha(i) is required to exceed the minimum
pressure head requirement of cooler E-i. When cooler E-i receives
the used water from cooler E-j, the flow rate and minimum pres- Table 1
Hot stream data.
sure head requirement of cooler E-i are equal to those of cooler
E-j, and the minimum pressure head requirement of branch-i is Stream Tin(°C) Tout(°C) Fcp (kW/h) h (W/m2 °C) Installation height
set to zero because branch-i does not exist at all. of coolers (m)
!
X
n 1 50 30 200 854 4
Hm þ HaðiÞ P 1 Zj; i  HðiÞ ð18Þ 2 50 40 150 743 9
j¼1 3 105 40 60 720 16
4 85 65 100 1352 32
In Eq. (19), n is the total number of coolers. The total number of aux- 5 90 55 80 450 19
iliary pumps should be less than the number of branches.
Pn Pn
i¼1 j¼1 Zi; j represents the number of series connections. The
number of branches equals the total number of coolers minus the Table 2
number of series connections. Physical properties and constraints of the case study.
X
n X
n X
n
Items Data
yi 6 n  Z i;j ð19Þ
j¼1 i¼1 j¼1 Minimum approach temperature difference 4Tmin 10 °C
Maximum allowable outlet temperature of cooler Tmax 55 °C
Eqs.(20–22) express the capital and operational costs of the Cooling water specific heat capacity Cpt 4.18 kJ/(kg °C)
pumps. Here, Cfpump, Cpump, and c are the parameters of the pump Film heat transfer coefficient hw 2.5 kW/(m2 °C)
capital cost and M is the mass flow rate. Density of cooling water q 995 kg/m3
Viscosity of cooling water mt 1  10-3 kg/m s
CC pump ¼ Cfpump þ C pump ðMDP=qÞc ð20Þ Viscosity correction factor Ut 1.05
Conductivity kt 0.6 W/m °C
Hm  ft  g Tube internal diameter Di 15.4 mm
OC mainpump ¼ eh ð21Þ Tube external diameter Di 19.1 mm
gpump
382 J. Ma et al. / Applied Thermal Engineering 125 (2017) 377–385

Table 3 is optimized. Then, the optimal parallel structure is used to com-

Economic data of the case study.
Items
Heat exchanger capital cost ($) 30000 + 1,000°A
Pump capital cost ($)
Plan lifetime
Electricity rate
Pump efficiency
Unit cost of water
Plant operation time
Interest rate
Annualized factor
treated as the optimal variables. The economic data of the case
study is presented in Table 3.
pare with the design obtained from the proposed method. In the
Cost data Remarks initial parallel structure, the optimized variables are the cooling
0.5
A in m2 water flow rates, contact areas, and pressure drops. The structure
2,000 + 5 (M4p/ 4p in Pa, q in kg/m3, of network is kept fixed. After optimizing the structure with the
q)0.68 M = kg/s mathematical model previously mentioned, the optimal parallel
5 years structure is derived, as presented in Fig. 3.
0.1 $/kWh
70%
In Fig. 3, the total mass flow rate is 137.28 kg/s, outlet temper-
2.0  10-6 $/kg ature is 44.75 °C, and total exchanger area is 1193.80 m2. The total
2.88  107 s/year annual cost of this configuration is $ 126,865.16. The main pump
15% pressure head is required to exceed the minimum pressure head
0.298
requirements of all the coolers, and it is 32.77 m. The pump oper-
ational cost is extremely high, and it is necessary to retrofit the
configuration with auxiliary pumps. There is no reusable cooling
water in the system; therefore, the cooling water consumption is
high. Consequently, it is necessary to switch the configuration into
the series–parallel type.

5.1. Original parallel configuration

In the initial configuration of cooling water system, the cooler
network is in parallel. According to the installation heights of the
coolers provided by Sun et al. [21], the corresponding space dia-
gram is assumed as shown in Fig. 2. The installation heights of
the coolers are varied, and there is only one main pump to trans-
port the cooling water from the cooling tower to the coolers.
To verify the effectiveness of the two-step sequential and simul-
taneous optimization methods, the original parallel configuration
5.2. Two-step sequential optimization
The two-step sequential method was proposed by Sun et al.
[21]. The first step in it is switching the parallel cooler network into
the series–parallel structure. Using the specified cooler network
and water flow rate, the pump network can be optimized. The aux-
iliary pumps are installed on the branch pipelines. To demonstrate
that the simultaneous optimization method outperforms the two-
step sequential method, the two-step sequential optimization was

E-4
Cooling tower

E-5

E-3

E-2

E-1
Main pump

Fig. 2. Initial parallel configuration space diagram.

E-1 36.96°C
56.39kg/s
30 50
548.88m2

E-2
38.12°C
19.80kg/s
40 50
108.01m2

E-3
137.28kg/s 26.65kg/s 55.00°C 44.75°C
20°C 40 105
213.28m2

Main pump E-4

55.00°C
13.67kg/s
65 85
61.61m2
E-5
20.75kg/s 38.12°C

55 90
202.01m2

Fig. 3. Initial parallel configuration two-dimensional diagram.

 J. Ma et al. / Applied Thermal Engineering 125 (2017) 377–385 383

Cooling tower
E-2 E-3
38.26kg/s 29.4 53.8
40 50 40 105
128.95m 2 272.25m2

E-1 E-4
115.83kg/s 57.85kg/s 36.5 44.8 49.32
20 30 2 50 65 85
539.65m 67.06m2

Main pump
E-5
19.72kg/s 53.9

55 90
206.76m2

Fig. 4. Optimal cooler network without auxiliary pumps.

Cooling tower
E-2 E-3
38.26kg/s 29.4 53.8
40 50 40 2 105
128.95m2 272.25m

E-1 E-4
115.83kg/s 57.85kg/s 44.8 49.32
20 30 2 50 65 85
539.65m 67.06m2
Auxi pump
Main pump
E-5
19.72kg/s 53.9

55 2 90
206.76m

Fig. 5. Optimal cooler network with auxiliary pumps.

conducted. The corresponding results are compared with the
results from the simultaneous optimization method.
First, the parallel configuration was optimized, Fig. 4 shows the
series–parallel configuration. It can be seen that coolers E-2 and E-
3 and coolers E-1 and E-4 are in series. When the cooler network
has a series–parallel structure, the cooling water consumption
decreases. The outlet temperature is 49.32 °C that is slightly higher
than the original parallel network. Although the total exchanger
area is larger, the water flow rate decreases. With the decreasing
water consumption, the cost of the series–parallel cooler network
is lower than that of the original parallel cooler network.
For a given flow rate, the cooler and pump networks can be
optimized. Fig. 5 displays the final configuration.
There is one auxiliary pump installed on the branch pipeline,
and the pressure head of that pump is 13.00 m. The pressure head
of main pump is 19.83 m. Decrease of the main pump pressure
head, decreases the pump operational cost significantly. The total
annual cost of this configuration is $ 109,896.37. In comparison
with the original structure, this configuration can save up to
13.4% of the total annual cost.
5.3. Simultaneous optimization
Two-step sequential method has certain limitations. When the
optimal cooler network is established, the pump network may
not be the most effective configuration. The intrinsic interactions

Cooling tower
E-4
13.67kg/s 55.0

65 85
61.61m2
Auxi pump

E-1 E-5
98.43kg/s 47.85kg/s 40.0 54.0 54.5
20 30 628.38m2 50 55
306.33m2
90

E-2 E-3
Ain pump 36.91kg/s 29.7 55.0

40 50 40 105
130.04m2 279.89m2

Fig. 6. Simultaneously optimized configurations of the pump and cooler networks.

384 J. Ma et al. / Applied Thermal Engineering 125 (2017) 377–385

E-4

E-5
Cooling tower

E-3

Auxi pump

E-2

E-1
Main pump

Fig. 7. Simultaneously optimized configuration space diagram.

between the pump and cooler networks require both to be opti-
mized simultaneously.
After optimizing the original structure with the simultaneous
method, the most optimal configuration was obtained, as shown
in Fig. 6.
Fig. 7 shows the simultaneously optimized configuration space
diagram.
Table 4 is employed to compare the physical data of the differ-
ent optimization methods.
In comparison with the two-step sequentially optimized struc-
ture, the total water flow rate reduces to 98.43 kg/s and the outlet
temperature increases to 54.5 °C. In addition, the cooler network
becomes different. In the two-step sequential structure, coolers
E-1 and E-4 are in series. However, coolers E-1 and E-5 are in series
in this structure, and the auxiliary pump only transports the cool-
ing water for cooler E-4. In the two-step sequential structure, the
auxiliary pump that has a pressure head of 13.00 m sends
38.26 kg/s cooling water to coolers E-1 and E-4. In the simultane-
ously optimized configuration, the auxiliary pump is only required
to send 13.67 kg/s cooling water to cooler E-4, while its pressure
head is still 13.00 m. The discharge rate of the auxiliary pump
decreases, so that both its capital and operational costs decrease.
Consideration of the installation height tends to decrease the water
flow rate, thereby saving the pumping cost. Particularly for the
cooler that is installed on a high platform, the flow rate of that
cooler tends to decrease drastically. When there is less cooling
water to cool down the hot stream for a fixed heat load and heat
transfer coefficient, the outlet temperature of the cooling water
increases, and the temperature difference between the hot and
cold streams decreases. For a lower temperature difference, more
contact area is required to achieve heat exchanging service. Conse-
quently, the capital cost of the coolers is increased owing to the lar-
ger contact areas. Coolers E-2 and E-3 are still in series; however,
the flow rate of this branch reduces from 38.26 kg/s to
36.91 kg/s, and cooler E-3 outlet temperature increases from
53.8 °C to 55.0 °C. After optimizing the pump and cooler networks
simultaneously, we determine that the highest cooler E-4 does not
connect with any cooler. This is because when the highest cooler
connects with other coolers, the water flow rate of this branch is
higher and the pressure head is based on the height of the highest
cooler. The pumping cost will be higher when the highest cooler
connects with other coolers. The outlet temperature of cooler E-4
is 44.8 °C in the two-step sequential structure, whereas now it is
55.0 °C. The main pump pressure head is 20.55 m that is slightly
higher than the previous main pump pressure head. This is because
the exchanger contact areas increase, causing the pressure drop of
the cooler network to increase. However, the total flow rate is less
than that of the previous structure flow rate; Consequently, the
main pump capital cost, pump operational cost, and cooling water
consumption decrease. Table 5 compares the economic data of the
different optimization methods.
Although the exchanger capital cost increases, the pump and
water costs reduce more. The total annual cost is $ 102,864.05 in
comparison with the two-step sequential method, and the

Table 4
Comparison of the physical data of the different optimization methods.

Original parallel optimization Two-step sequential optimization Simultaneous optimization

Total flow rate 137.28 kg/s 115.83 kg/s 98.43 kg/s
Cooling tower inlet temperature 44.75 °C 49.32 °C 54.50 °C
Cooling tower outlet temperature 20.00 °C 20.00 °C 20.00 °C
Total area of the exchangers 1193.8 m2 1214.68 m2 1406.25 m2
Pressure drop of the cooler network 7480.11 Pa 8,058.94 Pa 15,012.74 Pa
Main pump pressure head 32.77 m 19.83 m 20.55 m

Table 5
Comparison of the economic data of the different methods.

Original parallel optimization Two-step sequential optimization Simultaneous optimization

Cost of cooling water ($) 7907.22 6671.83 5669.45
Capital cost of exchangers ($) 64,686.51 64,859.85 68,108.66
Operational cost of main pump ($) 51,571.25 26,803.89 24,357.10
Operational cost of auxiliary pumps ($) 0.00 8423.73 1990.43
Capital cost of main pump ($) 2774.65 1992.13 1904.15
Capital cost of auxiliary pumps ($) 0.00 1231.47 834.25
Total annual cost ($) 126,865.16 109,896.37 102,864.05
 J. Ma et al. / Applied Thermal Engineering 125 (2017) 377–385
simultaneous optimization method saves 6.4% of the total annual
cost. In comparison with the original parallel structure, this config-
uration saves 18.9 % of the total annual cost.
6. Conclusion
In this study, the pump and cooler networks were optimized
simultaneously. Both the height and pressure drop of the cooler
were considered in the optimization of the pump network. A flex-
ible and practical series–parallel cooler network was employed.
The auxiliary pumps were installed on the branches to save the
pumping cost.
The interaction between the pump and cooler networks was
examined. When the pump and cooler networks are optimized
simultaneously, there was less cooling water in the system com-
pared with the other optimized configurations. In particular, for
the coolers installed on high platforms, the cooling water flow rate
tended to decrease drastically. When the highest cooler connected
with the other coolers, the flow rate and pumping cost were extre-
mely high. Thus, the probability of the highest cooler connecting to
other coolers on the simultaneous optimization of the pump and
cooler networks was low. Although, consuming less cooling water
requires more contact areas, savings in the pumping cost were
more. A case study was employed to verify the effectiveness of
our model. A comparison of the physical and economic results of
the three different configurations demonstrated that the simulta-
neously optimized model outperformed the other models. The pro-
posed model is predicted to save 18.9% and 6.4% of the total annual
cost in comparison with the original parallel structure and two-
step optimized model, respectively.
Acknowledgements
Support from the National Natural Science Foundation of China
(21576286) Science Foundation of China University of Petroleum,
Beijing (No. C201606) are gratefully acknowledged.
References
[1] D. Kaya, E.A. Yagmur, K.S. Yigit, F.C. Kilic, A.S. Eren, C. Celik, Energy efficiency in
pumps, Energy Convers. Manage. 49 (2008) 1662–1673.
[2] J.K. Kim, R. Smith, Cooling water system design, Chem. Eng. Sci. 56 (2001)
3641–3658.
385
[3] X. Feng, R. Shen, B. Wang, Recirculating cooling-water network with an
intermediate cooling-water main, Energ. Fuel. 19 (2005) 1723–1728.
[4] M.H. Panjeshahi, A. Ataei, M. Gharaie, R. Parand, Optimum design of cooling
water systems for energy and water conservation, Chem. Eng. Res. Design 87
(2009) 200–209.
[5] J.M. Ponce-Ortega, M. Serna-González, A. Jiménez-Gutiérrez, MINLP synthesis
of optimal cooling networks, Chem. Eng. Sci. 62 (2007) 5728–5735.
[6] Y. Wang, K.H. Chu, Z. Wang, Two-step methodology for retrofit design of
cooling water networks, Indust. Eng. Chem. research 53 (2014) 274–286.
[7] M. Picón-Núñez, A. Morales-Fuentes, E.E. Vázquez-Ramı’Rez, Effect of network
arrangement on the heat transfer area of cooling networks, Appl. Therm. Eng.
27 (2007) 2650–2656.
[8] M. Picón-Núñez, G.T. Polley, L. Canizalez-Dávalos, E.K. Tamakloe, Design of
coolers for use in an existing cooling water network, Appl. Therm. Eng. 43
(2012) 51–59.
[9] J.Y. Lee, C.L. Chen, T.L. Wen, D.K.S. Ng, D.C.Y. Foo, T.C. Wang, Synthesis and
design of chilled water networks using mathematical optimization, Appl.
Therm. Eng. 58 (2013) 638–649.
[10] V.K.A. Shankar, S. Umashankar, S. Paramasivam, N. Hanigovszki, A
comprehensive review on energy efficiency enhancement initiatives in
centrifugal pumping system, Appl. Energ. 181 (2016) 495–513.
[11] S.A. Tirmizi, P. Gandhidasan, S.M. Zubair, Performance analysis of a chilled
water system with various pumping schemes, Appl. Energ. 100 (2012) 238–
248.
[12] T. Westerlund, F. Pettersson, I.E. Grossmann, Optimization of pump
configurations as a MINLP problem, Comput. Chem. Eng. 18 (1994) 845–858.
[13] Z. Zhang, Y. Zeng, A. Kusiak, Minimizing pump energy in a wastewater
processing plant, Energy 47 (2012) 505–514.
[14] S. Frausto-Hernández, V. Rico-RamíRez, A. Jiménez-Gutiérrez, S. Hernández-
Castro, MINLP synthesis of heat exchanger networks considering pressure
drop effects, Comput. Chem. Eng. 27 (2003) 1143–1152.
[15] R.D. Souza, S. Khanam, B. Mohanty, Synthesis of heat exchanger network
considering pressure drop and layout of equipment exchanging heat, Energy
101 (2016) 484–495.
[16] F. Torkfar, A. Avami, A simultaneous methodology for the optimal design of
integrated water and energy networks considering pressure drops in process
industries, Process Safety Environment. Protect. 103 (2016) 442–454.
[17] S.W. Hung, J.K. Kim, Optimization of Water Systems with the Consideration of
Pressure Drop and Pumping, Indust. Eng. Chem. Res. 51 (2012) 848–859.
[18] K.V. Gololo, T. Majozi, Complex cooling water systems optimization with
pressure drop consideration, Indust. Eng. Chem. Res. 52 (2012) 7056–7065.
[19] H. Soltani, S. Shafiei, Heat exchanger networks retrofit with considering
pressure drop by coupling genetic algorithm with LP (linear programming)
and ILP (integer linear programming) methods, Fuel Energy Abstr. 36 (2011)
2381–2391.
[20] J. Sun, X. Feng, Y. Wang, C. Deng, K.H. Chu, Pump network optimization for a
cooling water system, Energy 67 (2014) 506–512.
[21] J. Sun, X. Feng, Y. Wang, Cooling-water system optimisation with a novel two-
step sequential method, Appl. Therm. Eng. 89 (2015) 1006–1013.
[22] J.M. Ponce-Ortega, M. Serna-González, A. Jiménez-Gutiérrez, Optimization
model for re-circulating cooling water systems, Comput. Chem. Eng. 34 (2010)
177–195.
[23] J.J.J. Chen, Comments on improvements on a replacement for the logarithmic
mean, Chem. Eng. Sci. 42 (1987) 2488–2489.